Month: June 2026

The “baby boom” generation, made up of Americans born from after World War II up into the early 1960s, is sometimes called the “pig in the python” of US demographics. The pig started moving into retirement age around 2010, when the earliest of the boomers hit age 65. At present, the tail end of the boomers (like me) are hitting retirement age. This shift has some well-known implications, like the rising financial pressures on Social Security and Medicare. But here, I want to focus on a different implication: With large numbers of retirees, and lower birthrates after the baby boom generation, it seems as if the supply of labor in the US economy should be growing at a relatively slow pace for the next decade or two. In addition, basic economics suggest that a slowdown in supply should tend to raise the price–in this case, raise the wage for young workers entering the job market. How likely is this scenario?

Steven Ruggles offers some calculations and insights in “The pig in the python: US decennial labor flows and economic opportunity, 1910–2040” (PNAS, May 15, 2026, vol. 123, #20, e2601716123). Here’s one illustrative figure. Let’s calculate the “net” entrants to the US labor force in each decade, and we will do the calculation as a share of the size of the pre-existing workforce in that decade.

During the high-immigration period at the start of the 20th century, net entrants to the labor force were in the range of 8-10% of the of the working-age population. After the clampdown on immigration in the 1920s and the miserable economic period of the 1930s, this dropped a bit, and the rises slightly again in the 1950. But then you see the big spike of new entrants into the US labor force in the 1960s, whihc is the boomers entering the labor force. After this boom, new entrants sag ove time. The level is currently down to below the Great Depression levels, and in the decades from 2030-2040, net entry into the labor force is projected to turn negative (with a combination of lower US birthrates and recent clampdowns in immigration to the US).

What are the implications for younger workers. Ruggles summarizes in this way:

We are on the verge of a fundamental transformation of the labor market. Baby boomers began retiring in large numbers during the 2010s, and immigration began declining in the same period. The number of births went down 17% between 2007 and 2024, which will contribute to a reduction in the number of new entrants to the labor force in the 2030s.

In the current moment of political turmoil, any economic predictions spanning the next 15 years are highly uncertain. We can predict with some confidence, however, the general demographic configuration of the 2040 working-age population: the births have already occurred, mortality change is ordinarily gradual, and a massive surge of immigration seems unlikely. Other factors may affect the prospects of new workers, including recessions, sudden shifts in federal policies, and long-running economic trends like globalization and mechanization. It is possible, as some prognosticators contend, that artificial intelligence will dramatically reduce the demand for workers. It is also possible—or even probable—that the magnitude of the coming demographic transformation may be great enough to swamp such economic changes.

Barring revolutionary changes in the economy, the drop in labor-market competition over the coming two decades will have profound consequences. There is likely to be an unprecedented shortfall of new workers, creating strong upward pressure on wages. … [W]e are already seeing signs of an uptick in the wages of young workers, and as the demographic shortage accelerates we may finally see real wages of the young exceed the levels of the early 1970s. Americans born in the 2020s might be the first cohort in a half century that earns significantly more than their parents did. Labor-force participation will probably increase as more workers are pulled into the market by higher wages, and some workers will postpone retirement. The decline in labor competition should be a boon for labor organizing. There will be increased incentives for automation of both manufacturing and services. We can expect reduced inequality, especially generational inequality. High demand for labor will create pressure to expand immigration. The unprecedented drop in employment competition may also have adverse consequences. Labor shortages may constrain economic growth, although there is some evidence to the contrary.

The movement of the boomer generation through education, the labor market, and now into retirement has been one of the dominant factors for the US economy for decades. But that period is, gradually, coming to an end.

The standard definition of “behavioral economics” is that it brought insights from psychology to bear on economic decision-making: that is, concepts like using rules-of-thumb in situations with limited information, valuing the present over the future, difficulties in perceiving probabilities accurately, habit formation, peer effects, and more. But the most recent edition of the Behavioral Economics Guide 2026, edited by Alain Samson, includes two essays suggesting that it’s time for additional steps. In a similar spirit, Ulrike Malmendier argues that it’s time to take personal experience effects into account; Isabelle Brocas argues for applying insights from the biological sciences.

In “Homo Experiens: Why Behavioral Economics Needs the Life Sciences,” Malmendier points to a body of research which finds (perhaps unsurprisingly to anyone but economists) that past experience influences behavior. For example, the stock-market returns that people have experienced during their lives have a strong effect on people’s likelihood of investing in the stock market. Those who have lived through periods of higher inflation are more likely to expect higher rates of future inflation. Even experts, well-acquainted and up-to-date with evidence, are not immune to these experience effects. In a study of members of the Federal Reserve Board of Governors, those who had lived through periods of higher inflation were less likely to favor lower interest rates in the future. The beliefs of fund managers and doctors are shaped by life experiences.

Malmendier suggests that economic decision-making should expand to consider homo experiens, as she puts it. She writes:

Yet, the behavioral economics revolution has stalled at a decisive point. The human element I appealed to above is still missing, or at least incomplete. To put it in stark terms, human behavior still seems rather robotic. Where previously, in the neoclassical model, it was the output of a perfectly programmed computer, in the behavioral economics model it became the output of a not-quite-so-perfect computer program, one prone to systematic bugs. And while the discovery of heuristics and biases was a genuine improvement, the resulting models remained mechanical. Once we identify the biases and model them, the program is assumed to run the same way for everyone, at every point in their lives. What you have lived through, what you have felt, what has happened to your body and brain along the way—none of that enters the model.

In practice, how might such a research agenda be carried out? A starting point is just to take life experience into account as a variable. But Malmendier has more in mind:

The implication is that the biological channels through which experiences affect us—stress hormones, neural rewiring, immune and endocrine responses—are real, measurable, and consequential for the economic outcomes we care about. And they are almost entirely absent from economic models, whether behavioral or not. … I believe that the next chapter of behavioral economics lies in a turn towards the life sciences—in taking seriously what neuroscience, psychiatry, medicine, and biology already know about human beings and how they are shaped by what they live through. This means, first, that economists should start collecting data on stress, emotions, hormones, and physical health alongside the economic variables we traditionally measure. So far, economists have rarely gathered data along these dimensions, and the standard sources of survey data—the ACS [American Community Survey], the Michigan Survey of Consumers, the Survey of Consumer Finances—contain few variables that are related. …

In another essay in the same volume, Isabelle Brocas takes a related by different angle in “Beyond Preferences: The Biological Foundations.” Her focus is that behavioral economics can often seem like a list of possible biases and heuristics, operating with a various domains of decision-making, but with little sense that the domains are interrelated. She offers a useful diagram. The peach-colored labels around the outside focus on areas often studied by behavioral economists. The green labels show the brain mechanisms behind these areas of decision-making. As the graph usefully illustrates, mechanims that affect one domain are likely to affect several domains.

Perceiving this system as a unified whole, rather than a one-off list of individualized bits and pieces, seems like a worthy goal. Brocas writes:

One of the [economics] discipline’s great strengths has been its ability to isolate dimensions of behavior. Risk preferences, intertemporal preferences, ambiguity attitudes, social preferences, strategic sophistication, and self-control have all generated productive literatures. Yet, this partitioning has also come at a cost. It has encouraged the idea that these domains correspond to separable psychological objects, when many may reflect overlapping underlying mechanisms.

Take risk and time. Economists often treat them as distinct dimensions of preference: one concerns uncertainty, the other delay. But that separation becomes less convincing once we think mechanistically. Both types of decisions recruit valuation, affective anticipation, future representation, attention to salient outcomes, learning from feedback, and inhibitory control. A person who is impulsive may also appear more willing to take risks, not because impatience and risk-loving are identical constructs but because both may be shaped by common patterns of reward sensitivity, limited future simulation, stress, or weaker self-regulation. Similar overlaps likely exist between strategic reasoning and working memory, between cooperation and emotional regulation, between trust and threat perception, and between consumer choice and attentional bottlenecks. The traditional language of preferences can therefore obscure as much as it reveals. It describes regularities at the level of outcomes, but it does not tell us whether those regularities arise from common mechanisms or genuinely distinct systems.

Brocas argues that fruitful work can be done by looking at biology, and perhaps especially work in neuroeconomics.

A biologically informed economics does not require economists to abandon preferences or reduce behavior to neurons and genes. Preferences remain useful reduced-form representations when behavior is stable and prediction is the objective. The point is different: when economists seek to explain heterogeneity, diagnose why interventions fail, or design policies that work across contexts and populations, it is often necessary to model the mechanisms that generate the outward pattern of choice. Biology can enter economics not only empirically, but also theoretically through models that derive observed behavior from underlying brain processes and resource constraints. This perspective clarifies what is malleable, when it is malleable, for whom, and through which policy levers. Taking biology seriously is therefore not a detour from economic theory—it is a step toward a more explanatory and more useful science of human behavior. …

Neuroeconomic work has been especially valuable here because it shows that value-based choice relies on a sequence of interacting computations rather than on isolated behavioral modules. The act of choosing involves representing the problem, assigning subjective values, selecting an action, experiencing an outcome, and updating behavior through learning (Rangel et al., 2008). This logic already pushes against the idea that risk, time, social choice, and self-control should be understood as self-contained domains.

I can barely keep up with the broad field of economics as it is. I frankly despair of becoming knowledgable in the neuroeconomics of brain-based decision-making, including “stress hormones, neural rewiring, immune and endocrine responses.” But I’m always intrugued by the work of economists like Malmendier and Brocas who are willing and able to make the effort.

Finance & Development has published a five-paper symposium on “Geoeconomics” (June 2026)., which is the idea that in a global economy, national security policy needs to be incorporated into economic policy-making. Christopher Clayton, Matteo Maggiori, and Jesse Schreger contribute “Understanding Geoeconomics in a Volatile World.” They write:

The academic study of geoeconomics dates most prominently to 1945, when economist Albert Hirschman published National Power and the Structure of Foreign Trade. In it, he examines how Nazi Germany had structured its economy to maximize leverage over its neighbors during the interwar period. He rejected the naive view that because trade is voluntary and mutually beneficial, it is geopolitically harmless. Benefits can be mutual, Hirschman argues, without being symmetrical. And asymmetry is how power builds. Since Hirschman’s time, economists have left the study of global power dynamics largely to political scientists and historians, who have led the development of this area of research. Though almost every economics student encounters the Herfindahl-Hirschman Index, few know it was invented to measure the economic power of nations, not firms. 

Clayton, Maggiori, and Schreger discuss some recent episodes of the use of economic sanctions, including those imposed on Russia, and draw some general lessons. For example:

Inputs are called choke points, or critical dependencies, if the hegemon controls a dominant market share of the input in the targeted economy and it is difficult to find alternatives to the hegemon’s inputs. For example, the US and its allies control an overwhelming share of global financial services, upward of 80 to 90 percent in many countries. Payment systems, settlement infrastructure, and dollar-denominated lending are basic inputs in a functioning economy. The lack of viable alternatives to the US financial infrastructure gives the country considerable geoeconomic power. Recently, it has wielded this power by imposing comprehensive financial sanctions on Iran and Russia, putting pressure on HSBC to disclose transactions linked to Huawei, and cutting Russian banks’ access to the SWIFT messaging system for international financial transactions.

However, there is a catch. The relationship between control over a sector and geoeconomic power is not linear; rather, power increases disproportionately as a hegemon approaches complete control. The difference between controlling 95 percent and 85 percent of an input is disproportionately large. At 95 percent, a target economy has almost no viable alternatives and must accept whatever terms the hegemon demands. At 85 percent, there is enough of an alternative to give the target meaningful options, and the hegemon’s leverage dissipates rapidly. …

Our work shows that there is a trade-off between gains from trade and economic security. The same mechanisms that are the classic foundations of the gains from trade—economies of scale and specialization—also generate economic dependence. The domestic alternatives that countries did not build up are poor substitutes for globally dominant inputs, such as Chinese manufacturing or US financial services and technology. This lack of alternatives leaves the countries exposed to coercion. As the global economy increasingly relies on goods and services that have strategic complementarities and economies of scale, these mechanisms are likely to increase in importance.

At one level, geoeconomics is just common sense: no one would want to see a nation’s security weakened to make some academic point about the merits of free trade. Thinking more seriously about diversificaition of global supply chains, and about interdependencies between nations, seems fully worthwhile. But as a matter of practical politics, claims about “national security” also deserve some skepticism. But when you run “national security” through the corridors of Washington, DC, you will find that it collects a lengthy wish-list of subsidies and tax breaks for industries, K-12 education, national health care, climate change, and government access to personal information and much more–are of which, amazingly enough, turn out to be core problems of “national security.” The “national security” label is no excuse for not considering whether the proposed policy is actually appropriate, the full costs of each policy, and the full range of alternative policy choices.

Jerome Powell was Chair of the Federal Reserve from February 2018 to May 2026. Of course, the Fed Chair does not have dictatorial power over setting monetary policy: instead, such policy is set by a vote of the 12-member Federal Open Market Committee, which in turn is made up of the seven DC-based members of the Fed Board of Governors and five presidents of the regional Federal Reserve banks (chosen on a rotating basis). However, the Chair does have, by tradition, position, and leadership, disproportionate influence over the monetary policies that are selected. How should Powell’s record as Chair be evaluated? Christina D. Romer  and David H. Romer discuss “An early retrospective on monetary policy in the Powell era” (Hutchins Center Working Paper #106, Brookings Institution, June 2026).

Rather than trying to do a month-by-month or meeting-by-meeting evaluation, Romer and Romer quite sensibly focus six main episodes in monetary policy during Powell’s tenure as Chair:

We begin our study with a selective chronology of monetary policy since 2018. We don’t attempt to provide a comprehensive blow-by-blow description of Federal Reserve actions. Rather, we focus on what we see as six relatively distinct policy episodes. These are: (1) the interest rate and balance sheet normalization in 2018 and early 2019; (2) the reversal of both these policies in mid- and late-2019; (3) the aggressive expansionary response to the COVID 19 pandemic in 2020 and early 2021; (4) continued loose policy in 2021 as inflation surged; (5) the rapid tightening in 2022 and 2023 to fight inflation; and (6) the interest rate cuts starting in mid-2024 and severe threats to Fed independence.

This table shows the dates of the six episodes, along with the two main policy tools that the Fed was using during this time: changes in the federal funds interest rate (the specific interest rate that is targetted by the Fed) and changes in Fed holdings of securities–US Treasury bonds and mortgage-backed securities.

Here’s a figure showing the path of the target for the federal fund interest rate and for Fed holdings of securities over time. Remember that coming out of the Great Recession of 2007-09, the Fed was doing everything it could to support the economy, including a target interest rate set barely above zero percent and a willingness to hold $4 trillion in securities. Thus, when Powell is named Chair by President Trump in 2018, there is already a goal underway to move interest rates and Fed holdings of securities back to a more usual level.

The first two monetary episodes mentioned by Romer and Romer thus involve continuing the process of raising interest rates and gradually reducing Fed holdings of securities. However, in mid-2019, the Fed decided on a mild reversal of both policies. Romer and Romer write:

Attempts to shrink the Fed’s balance sheet, however, led to a temporary loss of interest rate control and a return to balance sheet expansion. The directional shift from interest rate hikes to cuts in mid-2019 can be understood as a response to slightly worse economic forecasts and a few low inflation readings, but it perhaps missed the more fundamental fact that unemployment was at historic lows.

Monetary policy is hard, and hindsight is 20:20. But as the charts above suggest, the monetary policy choices in 2018-2019 were first to raise interest rates by 1 percentage point, and then to cut them by .75 percent points. With Fed holdings of securities, the first step was to reduce holdings by $593 billion, while the second step was to raise them by $253 billion. These changes are not enormous ones, and they are overshadowed by the dramatic challenges of monetary policy that soon followed during the pandemic.

But as the Romers write about this time: “We see an increasing focus among FOMC members on theories that inflation was relatively impervious to economic conditions …” In other words, there was a belief that interest rates should just stay low, even when unemployment rates were also very low, because inflation was permanently under control. My own view is that these beliefs did not serve the Fed well during the period after the pandemic recession, and in some ways set the stage for Fed confrontations with the Trump administration in the last 18 months or so.

The pandemic recession hits in 2020. The Fed pulls out all the stops. Interest rates drop again to near-zero, and Fed holdings the US Treaury securities rises dramatically. There are also several severe strains in financial markets during this time, including a near-meltdown of the global market for US Treasury bonds in March 2020. Behind the scenes, “the Federal Reserve took aggressive actions to stabilize financial markets and ensure smooth market functioning.” This period (the third episode in the Romer and Romer chronology) is where Powell’s record as chair truly shines. The economic consequences of the pandemic are severe, but could have been so much more severe and longer-lasting without the Fed actions.

Conversely, the period right after the pandemic is the most controversial time for Powell’s legacy. Inflation rises, and the Fed was slow to respond, in part because members of the Fed continued to believe “that inflation was relatively impervious to a very strong labor market,” and that the surge of inflation would be only temporary. Romer and Romer note: “Mistaken forecasts that inflation would disappear quickly also factored into their policy choices. However, given that even the mistaken forecasts had inflation above target for an extended period and unemployment well below the natural rate, the decision to leave the funds rate unchanged for so long does not appear reasonable and had unfortunate consequences.”

In the fifth stage, the Fed finally acted to raise interest rates and reduce inflation, but the end result is that inflation remained above the target of 2% annual inflation since then. About this episode, Romer and Romer write:

It is very possible that the forceful moves and clear communication played a role in taming inflation, and did so with historically modest damage to the labor market. The gradual reduction in interest rates starting in mid-2024 was motivated by the Fed’s forecasts that it was closing in on both its inflation and maximum employment targets, and so could move toward its estimate of the neutral rate. However, similar to what we find for the 2019 cuts and the 2021 inflation forbearance, the FOMC arguably lost some sight of the bigger picture. In this case, the most important missing fundamental was simply that inflation had been well above target for years.

The final episode is the gradual interest rate cuts since 2024 and the pressure on Powell and the Fed from the Trump administration to cut rates faster. On one side, Powell has offered a strong defense of the independence of the Federal Reserve, despite sustained and very public personal and legal attacks. For this, he deserves considerable credit. But on the other side, the inflation rate has remained above the 2% annual target set by the Fed for several years now, while unemployment rates have been low by historical standards. Rather than keep interest rates at a higher level to drive out the remainder of inflation, the Fed has, albeit in a halting and episodic way, brought down interest rates and again started expanding Fed securities holdings.

Romer and Romer make no secret that their view is highly sympathetic to Powell; indeed, they note up front that they view him as a “hero.” My personal view toward Powell’s record is positive, but less so than that of the authors. My own sense is that Powell deserved considerable credit both for Fed actions to support the economy during the worst of the pandemic recession and also for his robust defense of Federal Reserve independence from President Trump. On the other side, it seems to me that the Fed under Powell has consistently leaned toward a belief that interest rates could be reduced or stay low with little or no risk of higher inflation: this was true in the 2019, in the aftermath of the pandemic recession, and again in the last 18 months or so. Moreover, one after-effect of the pandemic recession is that Fed holdings of securities, which were already viewed as too high when Powell assumed the Chair, are markedly higher now. Thus, in sustaining the inflation-fighting credibility of the Fed and also in how the Fed has become intertwined with US budget deficits by holding larger amounts of US Treasury securities, Powell’s two terms as Chair of the Fed have left some real challenges for his successor.

Gross domestic product measures the economic transaction in an economy, according to quantities bought and sold and market prices. But does this method work for AI? Anton Korinek and Patrick McKelvey work through the question in “Where is AI in GDP Statistics (Peterson Institute for International Economics, May 2026, both a readable Policy Brief and an underlying research paper are available).

Here’s the basic issue. The authors calculate that the total amount currently spent on AI computing power is (roughly, depending on underlying assumptions) about $250 billion per year. This measures the AI bought and sold in the market, and thus is clearly part of GDP.

However, the AI chips are becoming more efficient:”As chips became more efficient, each dollar of compute spending bought more physical computing capacity. Measured in H100 equivalent units, US AI computing capacity grew at more than 200 percent per year, outpacing nominal spending.” Also, the AI algorithms are becoming more efficient, so that the quantity of computing power needed to achieve a fixed level of AI has been falling by about two-thirds per year. Put these together, and the quality-adjusted increase in AI is more than 2000% (that is, a twenty-fold increase) per year.

So here’s a situation where the amount spent on AI and captured by GDP is rising quickly, but the actual capabilities of that AI spending are rising much, much faster. How should measures of the size of the economy deal with this situation?

The question isn’t totally new. Government statisticians have for some years now used “hedonic” adjustment. The idea is to define a good, like a computer or a television or a car, as a set of characteristics. The qualities of these characteristics are improving over time. Thus, when we say whether the “price of a computer” or the “price of a car” has changed, we want to do an apples-to-apples comparison holding the characteristics of that product constant. Thus, the Consumer Price Index measure for computers and related goods looks like this:

Just to be clear, this graph doesn’t mean that the amount you personally paid out of pocket for a computer fell by three-quarters from 2005 to 2018. It means that the price of buying a computer with the same characteristics as a standard 2005 computer fell by three-quarters by 2018–but of course, few of us in 2018 would have wanted to buy that version of the computer from 2005, even if it was available. The authors explain:

That AI’s growing footprint is so faintly visible in national GDP statistics has, in part, a straightforward accounting explanation. Nominal AI revenues grow only moderately because per-unit prices for any given level of AI capability fall almost as fast as quality-adjusted output rises. In the semiconductor industry, this pattern played out for decades: each generation of chips was dramatically cheaper per unit of performance than the last, so the semiconductor share of GDP remained modest even as quality-adjusted output expanded enormously.

This kind of hedonic adjustment could readily be used for AI as well, but for purposes of measuring GDP, it has limitations as well. GDP is a measure of market transactions, which happen when the technology behind supply-side production of goods and services meets the preferences and tastes behind demand-side purchases of those goods and services. Hedonic adjustments focus on adjusting for changes in the quality of what is produced, but don’t directly take into account what value buyers may place on these quality changes.

For example, perhaps the speed and power of AI tools have considerable value to users up to some point, but then additional speed and power have diminishing marginal value beyond that point. Just because quality-adjusted increase in output of AI has risen by 2000% per year in the last couple of years doesn’t mean that the value of AI tools to users has risen by 2000% per year.

Beyond the details of questions like these, Korinek and McKelvey are making a broader point: sensible thinking about the economic effects of the emerging AI tools needs to be rooted in data, but economic data about a very fast-changing industry can be hard to collect and can involve delicate issues of interpretation.

Consider a person who is convicted of a misdemeanor. It is their first offense. Should that person be punished to the full extent of the law, or dealt with more leniently?

Depending on one’s prejudices, a theoretical case can be made for either approach. Lenient treatment has a risk of causing the offender, along with anyone who hears about the lenient treatment, to believe that penalties for transgression are low or nonexistent, and in this way may encourage future transgressions. Harsh treatment, including jail time, has a risk of leading to additional consequences like loss of a job and restricted future employment possibilities, as well as exposing the first-time offender to seasoned recidivists.

On average and for the population as a whole, what does the empirical evidence say? Jennifer Doleac presents some evidence in “What Becomes of Second Chances?” (Behavioral Scientist, March 24, 2026). She points to a study that she has conducted with Amanda Agan and Anna Harvey based on data from Suffolk County, Massachusetts–where Boston is located. She writes: “In Suffolk County, once police make an arrest or issue a summons, and then determine that probable cause exists for the charge, the case goes to an arraignment hearing. In that hearing, an assistant district attorney (ADA) representing the government decides whether to pursue the charges or dismiss the case. They are essentially deciding whether they think the case is a good use of prosecutors’ time. This is the decision we were interested in. What if more cases were dismissed up front? Would that lead to more recidivism, or less?”

Their methodology relies on an underlying fact about this decision-making process. There are a bunch of assistant district attorneys. The nonviolent misdemeanor cases that are the focus of this studey are assigned to them pretty much at random, and given the volume of cases, the ADAs have limited time to make decisions about them. However, some of the ADA’s tend to be more strict, while others tend to be more lenient. To put it more bluntly, a given case is more likely to be either pursued or dismissed as a result of the random decision about which ADA gets the case.

From a standpoint of justice, the idea that the outcome has an element of randomness seems objectionable. From the standpoint of a researcher, it’s catnip. Using a statistical method (for the stat-minded, it’s a kind of instrumental variable called a leniency design), researchers can look at whether a higher number of cases pursued as a result of randomly being assigned to stricter ADAs (or equivalently, the higher number of cases dismissed as a result of being assigned to more lenient ADAs) affects later behavior. Doleac writes:

It turns out that leniency at this early stage—having your case dismissed rather than pursuing prosecution—reduced the likelihood of showing up in court again with new charges by 53 percent, and it reduced the number of future charges by 60 percent. The effects were larger for first-time defendants—those with no prior arrest or conviction on their record.

In other words, for nonviolent misdemeanor cases with no prior arrest or conviction, leniency works in the most practical and empirical sense. For those with prior arrests and/or convictions, leniency becomes less likely to work.

What about nonviolent felony cases, like burglary and car theft? Here, Doleac discusses a  study by Michael Mueller-Smith and Kevin Schnepel using data from Harris County, Texas, which includes the city of Houston. They were able to find two sources of underlying randomness.

On September 1, 1994, a Texas law went into effect that made it much less likely that prosecutors would offer “deferred adjudication”–basically, allowing the accused to go on probation for a period (say, six months) rather put off being tried for a nonviolent felony, and if the person had no further problems with law enforcement during that time, the felony charge would be reduced. Doleac writes:

This created the first natural experiment. The date of the policy change—September 1, 1994—sorted defendants into treatment and control groups, as if at random, based on the date of their offense. Nothing else changed at that date. The only difference between these defendants was whether they got a second chance to avoid a felony conviction. It turns out this second chance was very helpful. First-timers who got lucky and received a deferred adjudication committed fewer crimes going forward. They were 31 percentage points less likely to be convicted of any new crime over the next ten years—a 44 percent reduction compared with the control group.

The other natural experiment involved jail overcrowding in Houston. There was a referendum to expand jail space, but that referendum unexpectedly failed, which meant that a much larger share of those charged with nonviolent felonies were

Again, this set up a beautiful natural experiment. Mueller-Smith and Schnepel could compare defendants sentenced on either side of the election on November 6, 2007. The only difference between those sentenced before and after this date was that those sentenced after were much more likely to avoid a conviction. This difference wasn’t because of underlying differences between these defendants or their cases; it was because of the failed ballot initiative. This gave the researchers confidence that any future differences in recidivism or employment would be due to the diversion decision and not to something else about those defendants. Just as in 1994, there were big benefits to greater leniency. As the likelihood of diversion suddenly increased, the likelihood of new, future convictions fell, by 26 percentage points (46 percent). This is a dramatic change. Nearly half of the first-time offenders who would have committed another crime in the future if they’d been prosecuted and convicted as usual cleaned up their acts and avoided future crime when their cases were dropped or they received a deferred adjudication. 

This specific evidence on benefits of leniency has its limits. For example, it does not say that police activity or arrests should be reduced. As Doleac points out of those who received lenient treatment for nonviolent misdemeanors: “That person had likely been arrested and booked in jail, and had to show up in court for that initial hearing. This might mean taking time off work, and it certainly meant worrying about what might happen during the hearing. All this isn’t nothing—it is an inconvenience at best and a costly and stressful event at worst.” in addition, the focus of this evidence is on nonviolent misdemeanors and felonies, as well as on first-time offenders. Indeed, one additional practical reason for leniency in such cases is to focus the limited resources of the criminal justice system on repeat offenders and violent crimes.